Preoperative MR Image prediction of breast cancer based on self-supervised learning

Breast cancer is a prevalent cancer type among women worldwide, which contributes to significant mortality and morbidity rates. Early detection of breast cancer plays a crucial role in improving patients’ chances of survival. In this study, we utilized a publicly available dataset consisting of Magnetic Resonance (MR) images from 1,480 individuals. Leveraging deep learning technology, we successfully developed an accurate automatic detection system for breast cancer in MR images. To address the challenge of limited medical datasets, we propose a model based on self-supervised learning. This approach allows us to effectively overcome the limitations associated with small-scale datasets. Our experimental results demonstrate that our proposed model significantly enhances the predictive performance of breast cancer detection. Notably, our model achieved a remarkable six percentage point improvement in terms of AUC (reaching 0.95), when compared to the baseline model. The outcomes of this research present a promising step towards automated and precise prediction of breast cancer. The application of our model can aid in early-stage breast cancer detection, enabling timely treatment planning and ultimately enhancing patient survival rates.